I have two datacenter, dc1 3 nodes, dc2 3 nodes. For some reason, dc1 3 nodes are sync and have the latest data.
I run "nodetool repair -dcpar -full " on the first node of dc2. But after the repair, from Cassandra system log, repair for all range successful. But only node1 and node3 on standby got sync. node2 still missing the data.
Why this should be happened? any procedure needed before nodetool repair?
It's impossible for us to know what happened without actually analysing diagnostic data from your cluster. You will need to analyse the logs yourself to determine whether there was an issue with the repairs.
My recommendation is to run a rolling partitioner range repair (-pr) one node at a time:
$ nodetool repair -pr -- ks_name
A partitioner range repair is the most efficient way to run repairs and it minimises the impact on your cluster. Cheers!
5 People are following this question.
